I build evidence-centered systems — the kind that observe the real world, keep the evidence honest, and help a human make a better decision.
By day — a senior leader at a Big Four consulting practice, running hands-on factory-floor engagements (demonstrations, not slideware) across manufacturing modernization, supply-chain resilience, OT cybersecurity, and the defense-industrial base, energetics, and critical minerals.
After hours — I'm a maker at heart: a bench of soldering irons, logic analyzers, microcontrollers, and more breakout boards than I'll admit to. I design sensor-driven instruments end to end — selecting the silicon, writing the firmware, calibrating against reference sources, and building the software that turns raw readings into evidence — under one rule: raw truth over comforting lies.
Sprout (MIT) — a local-first plant-monitoring platform that refuses to lie. Real soil, real ESP32s, and the raw signal plus a band you calibrate yourself — never a fabricated "62% moisture." Honest status, in the repo and here: monitoring and calibrated bands run today on a live 8-sensor fleet; autonomous watering is roadmap, gated behind open engineering issues.
The way it's built matters as much as what it does: the first 18 days of history hold 528 commits, 433 merged PRs, 1,170 CI runs, and 31 architecture decision records — every number regenerable from the repository itself.
- C++ (embedded) — ESP32/Arduino-class and STM32 Nucleo firmware: sensor drivers, calibration routines, illuminator control, serial protocols.
- Python — data pipelines with schema-validated ingest, quality gates, test suites, analysis tooling.
- JavaScript — vanilla JS (typed with JSDoc comments) for instrument UIs and evidence dashboards.
Sprout is the newest of a couple dozen projects — same discipline, different domains: air-quality sensing and electronic-nose experiments, robots and autonomous rovers, spectral illumination and classification instruments spanning UVA illumination up through 1050 nm near-infrared — the upper edge of silicon-based sensing — and supply-chain evidence tooling. A few of these may follow Sprout into the open when they're ready.
A lot of my after-hours throughput comes from AI coding agents — and I treat that as an engineering problem, not a shortcut. I mix multiple platforms and models per task, coordinate them in role-specialized worktree lanes, and hold the output to the same change control I'd expect from a human team: PR-only merges behind CI, architecture decisions recorded while they're made, and a two-stage gate where an independent agent verifies acceptance criteria against source, history, and CI before anything merges. The interesting work is the harness itself — the guardrails, the verification loops, and the process that turns fast tools into trustworthy results. Across the private bench, that harness has moved ~2,500 commits through 645 gate-checked pull requests and 3,200+ CI runs in the last year. One instrument shipped this way carries a security posture aligned to NIST SSDF and DoD DevSecOps reference patterns: SHA-pinned actions, SBOM generation, secret scanning, locked coverage gates.
- Truth has a chain. Every reading is a value, from a source, at a time, with a confidence — sensor data, supply-chain evidence, and AI output alike.
- Systems degrade; design for it. Offline-first, honest under failure, honest about what it doesn't know.
- AI is a layer, not a source. It assists discovery and reasoning; it never replaces provenance, and a human stays in the loop.
Elsewhere — vkhogue.com · LinkedIn · always open to good conversations about AI strategy, industrial modernization, defense-industrial resilience, and critical materials.